One frame of real gallbladder surgery turned into 3D with Depth Anything 3 + SAM 3.1
MaziyarPanahi · x · 2026-11-07
- Developer @MaziyarPanahi took a single flat frame from a real gallbladder surgery and used Depth Anything 3 (ByteDance Seed, Apache-2.0) to turn it into a 3D scene on one GPU — effectively "flying inside" the operation.
- He then chained SAM 3.1 (Meta) on top: two words of prompting outline the gallbladder or liver and locate instruments.
- DA3 Metric Large specs: 0.35B-param monocular metric depth model supporting relative/metric depth and sky segmentation; multiply depth by focal length for real-world scale; exports glb/npz/ply/gsply formats.
- Licensing differs: DA3 is Apache-2.0, SAM 3.1 uses the SAM License, and the Endoscapes2023 dataset is CC BY-NC-SA 4.0 (non-commercial). Author notes it's research only, not medical advice.
Related event: Depth Anything 3 Turns Single Surgery Frame into 3D Scene(2 posts)→
More from Multimodal
- AI-generated vintage patent blueprints of biplanes and airships look stunningly detailed — azed_ai · 2026-11-08
- Filmmaker shares full workflow for a 1:42 AI dialogue scene: master voice track + timestamps — DavidmComfort · 2026-11-08
- Werewolf Image Generated Locally with Krea2 and MiniMax — -Ghosa- · 2026-11-08
- Midjourney 'Childhood Legends: Samurai Edition' reimagines Speed Racer & more — LudovicCreator · 2026-11-08
- LayerShop Adds Animated GIF Support for GPT Image 2.6 Generations — HonestTicket6013 · 2026-11-08
- Cyberpunk short film made with no gen-video: Opus wrote Blender scripts for everything — toni_btrain · 2026-11-08